Loading...

Design and Simulation of a Reinforcement Learning Based Power Management Strategy for Series-Parallel Hybrid Electric Vehicles

Ahmadian, Saeed | 2016

850 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 48059 (08)
  4. University: Sharif University of Technology
  5. Department: Mechanical Engineering
  6. Advisor(s): Alasty, Aria; Salarieh, Hassan
  7. Abstract:
  8. Growing consumer expectations, legislation pushing for lower emissions, increasing fuel prices, and the fact that petroleum is a finite resource are factors leading to groundbreaking changes in the automotive industry. Design and production of vehicles that use the lowest fuel are the most effective ways to reduce pollution. The use of both ICE and EM increases the complexity of HEV power management and advanced power management policy is required for achieving higher performance and lower fuel consumption. In this study, at first the benefits of series-parallel configuration are discussed comparing to other configuration. According to this, the present thesis provides a control strategy for series-parallel hybrid vehicles. Because of the complexity of this configuration, energy management of these vehicles is much more complex than others. In order to design an energy management strategy, the fuel consumption and the battery life are balanced that leads to lower fuel consumption and optimal performance of the battery. This was done using two methods. The first method uses the Particle Swarm Optimization algorithm (PSO) to optimize the rule base control strategy implemented on the series-parallel hybrid vehicle in ADVISOR software. The purpose of optimization is to reduce fuel consumption and increase the battery life. The second method is the use of reinforcement learning techniques (RL). This section aims at minimizing the HEV fuel consumption and increasing the battery life over any driving cycle (without prior knowledge of the cycle) by using a reinforcement learning technique. This is in clear contrast to the prior works, which require deterministic or stochastic knowledge of the driving cycles. In addition, the proposed reinforcement learning technique enables us to (partially) avoid reliance on complex HEV modeling while coping with driver specific behaviors. Inputs for the agent are the power demand of the vehicle, the level of battery charge and vehicle speed. Controller provides power requested from the battery as the output. Simulation results indicate that the proposed reinforcement learning based control is very effective for HEVs energy management
  9. Keywords:
  10. Reinforcement Learning ; Optimal Control ; Hybrid Electric Vehicle (HEV) ; Series-Parallel System ; Control Strategy ; Particles Swarm Optimization (PSO) ; Energy Management ; Series-Parallel Configuration ; Buttery Life

 Digital Object List

 Bookmark

No TOC